, SEMANTIC ANALYSIS (XU ET AL., HOVY ET AL., BETWEEN Q AND S. HOW...

2002), semantic analysis (Xu et al., Hovy et al.,

between Q and S. However, as research in QA

2001; Moldovan et al, 2002), and reasoning

demonstrates, word-overlap is not a good enough

(Moldovan et al, 2002). They access external

metric for determining whether a sentence contains

resources such as the WordNet (Hovy et al., 2001,

the answer to a question. Consider, for example,

Pasca and Harabagiu, 2001, Prager et al., 2001),

the question “Who is the leader of France?” The

the web (Brill et al., 2001), structured, and semi-

sentence “Henri Hadjenberg, who is the leader of

structured databases (Katz et al., 2001; Lin, 2002;

France’s Jewish community, endorsed confronting

Clarke, 2001). They contain feedback loops,

the specter of the Vichy past” overlaps with all

ranking, and re-ranking modules. Given their

question terms, but it does not contain the correct

complexity, it is often difficult (and sometimes

answer; while the sentence “Bush later met with

French President Jacques Chirac” does not overlap

with any question term, but it does contain the

In Section 2, we first present the noisy-channel

correct answer.

model that we propose for this task. In Section 3,

To circumvent this limitation of word-based

we describe how we generate training examples. In